Size, Age, and the Performance Life Cycle of Hedge Funds * Chao Gao, Tim Haight, and Chengdong Yin September 2018 Abstract This paper examines the performance life cycle of hedge funds. Small funds outperform large funds and small funds maintain good performance over time. One possible explanation for these effects is that expected management fees increasingly outweigh expected incentive fees when funds grow larger over their life cycle. Aside from size, performance life cycle patterns do not vary significantly with a host of fund- and family-level characteristics. Our results suggest that fund growth over time drives performance declines over a hedge fund s life cycle and that performance persistence is more achievable when funds stay small. Key Words: Hedge Funds, Performance Life Cycle, Fund Size, Fund Age. JEL Classification: G23. * We thank Lu Zheng for helpful comments and suggestions. All remaining errors are ours. Krannert School of Management, Purdue University. Email: gao202@purdue.edu. College of Business Administration, Loyola Marymount University. Email: thaight@lmu.edu. Corresponding author at: Krannert School of Management, Purdue University. 403 W. State Street, West Lafayette, IN 47907. Tel.: +1 (765) 494-4431. Email: yin80@purdue.edu.
How to select hedge funds with superior performance is one of the most extensively studied questions in the literature. While this task has always been important in the money management industry, it has been especially challenging for hedge funds in recent years, as hedge fund performance has been lackluster since the 2008-2009 financial crisis. In broad terms, the literature tackles the fund selection question from two different angles. One strand of the literature provides evidence on cross-sectional relations between fund performance and fund characteristics such as fund size, fund age, compensation contracts, and share restrictions. 1 Another strand of the literature examines whether hedge fund performance persists over time. 2 While there is scant research connecting these two strands, hedge fund investors are likely to consider characteristic predictors and performance persistence jointly. In particular, investors should be interested not only in identifying funds with superior performance in the cross section, but also in identifying how long superior performance can last, especially given hedge funds high minimum investment requirements and long share restriction periods. This study connects the two strands of literature by examining the performance life cycle of hedge funds with various characteristics. By performance life cycle, we mean performance at different stages of a hedge fund s existence. The performance life cycle approach offers several advantages over conventional approaches used in prior literature. First, the performance life cycle approach enables analyses of performance persistence over multiple periods, which is rare in the persistence literature. The literature commonly examines persistence over two consecutive periods using portfolio approaches and panel regressions, but the performance life cycle approach allows us to study how hedge fund performance evolves from a multi-period time series perspective. Second, the performance life cycle approach enables analyses of the first few years of a fund s 1 See Ackermann, McEnally, and Ravenscraft (1999), Liang (1999), Brown, Goetzmann, and Park (2001), Naik, Ramadorai, and Stromqvist (2007), Aragon (2007), Jones (2007), Agarwal, Daniel, and Naik (2009), Getmansky (2012), Schaub and Schmid (2013), Teo (2013), Aiken, Clifford, and Ellis (2015), and Yin (2016), among others. 2 See Eling (2009) for a review of earlier studies. See also Kosowski, Naik, and Teo (2007), Jagannathan, Malakhov, and Novikov (2010), and Ammann, Huber, and Schmid (2013), among others. 1
performance record, which is also rare in the persistence literature. 3 Because hedge funds typically have short lives, and because many successful funds maintain performance by closing themselves off to new investors, identifying performance persistence early in a hedge fund s life can be highly valuable for investors. Third, the performance life cycle approach enables performance comparisons of funds with different characteristics at different stages of their life cycles. Thus, the performance life cycle approach allows us to examine whether certain types of funds are more likely to maintain superior performance over time. One study that is closely related to ours is Aggarwal and Jorion (2010). They find that hedge fund performance peaks during the first few years of a fund s life, but declines thereafter at an average rate of 42 basis points per year. While these findings suggest that hedge fund performance declines with age (on average), the authors do not examine what drives performance declines with age, nor its connection with other fund characteristics. Thus, their findings may have limited implications for investors. Another related study is Boyson (2008), who examines hedge fund performance persistence by sorting funds based on fund size, fund age, and fund past performance. Like other studies in the persistence literature, Boyson (2008) only examines persistence over two consecutive periods, and persistence is not examined during the first few years of a fund s life. In addition, the author does not examine whether fund characteristics facilitate performance persistence independent of the level of past performance. While we add to the literature by examining how the performance life cycle of a hedge fund varies with its fund- and family-level characteristics, we are particularly interested in examining how fund age and fund size affect the performance life cycle. Fund age is naturally associated with the performance life cycle. However, prior studies generally limit their analysis of age effects to using age as a sorting variable in portfolio analysis or as a control variable in panel regressions. Moreover, Jones (2007) and Aggarwal and Jorion (2010) find that younger funds outperform older funds, but as noted earlier, the mechanism driving the age-performance relation 3 The persistence literature commonly uses fund performance over an evaluation period of 2 to 3 years to predict fund performance over the next period (one quarter or up to a few years in the future). Thus, there is very little evidence on performance persistence in the first 2 to 3 years of a hedge fund s performance record. 2
is unclear, so it is not well understood why we observe performance declines with age. Fund size may affect fund performance because of diseconomies of scale, a phenomenon that is well known to academics and practitioners. 4 Berk and Green (2004) develop a model in which good performance delivered by skilled managers attracts capital inflows, but the resulting fund growth leads to performance declines over time and a lack of performance persistence. Following Berk and Green (2004), several empirical studies provide evidence consistent with scale diseconomies in the hedge fund industry (e.g., Naik, Ramadorai, and Stromqvist (2007), Teo (2009), Getmansky (2012), and Yin (2016)). Nevertheless, most of these studies document scale diseconomies using panel data sets and thus offer limited insights about how size affects performance at different stages of a hedge fund s life cycle. By exploring size effects on the time-series dimension, we are able to investigate whether fund growth over time contributes to performance declines over a hedge fund s life cycle and we can assess whether size facilitates performance persistence in multi-year settings. We start with an analysis of fund performance over time. We collect data from the Lipper TASS and the HFR databases, and our main sample consists of non-backfilled funds that have at least five years of data. Non-backfilled funds are funds whose add dates are no more than 6 months after their inception dates. Aggarwal and Jorion (2010) discuss the importance of mitigating backfill bias when looking at age effects on performance, as backfilled funds are more likely to voluntarily report past performance when it is good. In addition, we require funds to have at least five years of data to mitigate a potential bias caused by funds that fail when they are young. Young failures may result from a variety of factors outside of a hedge fund s immediate control, such as adverse shocks to the industry (e.g., financial crises) or bad luck. Inclusion of young failures could drive a negative age-performance relation even when there is no age effect because performance is likely to deteriorate prior to liquidation. 5 Following Aggarwal and Jorion (2010), we group fund- 4 See Oksana Patron, Smaller Hedge Funds Firms are Doing Well, July 9, 2017, Sophie Baker, Smaller Hedge Funds are Able to Turn a Profit with Less than $100 Million AUM Survey, 2017, and Vishesh Kumar, Emerging Hedge Funds Outshine Established Peers as Investors Revisit Asset Class, July 10, 2017. 5 Our results are robust to other sample selection requirements as shown in Section IV. 3
month observations by event time, with the first event month being the first month performance data is available. Consistent with the literature, we find that fund performance declines with age. After documenting declining performance with age, we then investigate whether the ageperformance relation is driven by funds with certain fund- and family-level characteristics. For example, it is possible that funds with lockup periods are more likely to generate strong performance when they are young because they can invest in less liquid assets. Moreover, there is evidence that fund families have strong incentives to boost performance in their flagship funds to attract capital for the other funds in their family. After performing subset analysis, we continue to observe similar performance patterns across funds with varying characteristics. Next, we investigate how the performance life cycle varies with fund size. While prior studies examining hedge fund performance document diseconomies of scale in the cross section, we employ an event time approach to examine whether and to what extent scale diseconomies drive performance declines over a hedge fund s life cycle. In a preliminary analysis, we partition hedge funds in our main sample into three size groups at the beginning of each event year, and we observe average monthly portfolio performance over the event year. Results of this analysis indicate that the small group significantly outperforms the large group throughout much of the performance life cycle and that small group performance is quite consistent over time. One drawback to the portfolio approach is that portfolio membership varies across event years as fund size changes over time. To better examine the life cycle of hedge fund performance, we use modified Fama-MacBeth regressions where, for each fund, we run time-series regressions of performance on size and age, and we take cross-sectional averages of coefficients to assess size and age effects. We find that the cross-sectional average coefficient on fund size is negative and statistically significant while the age coefficient is marginally negative at best. Moreover, when we add further controls for fund family characteristics and capital flows, the age effect becomes statistically indistinguishable from zero. These results suggest that diseconomies of scale significantly drive performance declines over a hedge fund s life cycle. Since our main sample excludes backfilled funds and funds with less than five years of data, it is not clear whether investors can profitably exploit the effects that we document. For 4
example, small and young funds in our sample exclude those that fail within the first five years, but investors in real time cannot observe which small and young funds will succeed and which will fail. Thus, to provide evidence on the exploitability of the effects that we document, we expand our sample to include backfilled funds and non-backfilled funds with less than five years of data, and we repeat our portfolio analysis using nine (3x3) portfolios sorted independently on size and age at the beginning of each calendar year. Our results indicate that small funds outperform large funds in all three age groups. By contrast, we do not find strongly significant differences in performance between young funds and old funds. Thus, our analysis suggests that investors are more likely to earn higher returns by investing in small hedge funds. In addition, we also examine the frequency with which each size-age portfolio generates winning performance by ranking the performance of each portfolio in each calendar year. Our results indicate that portfolios that include small hedge funds are more likely to be in the winner group (i.e., top tericle of performance) and less likely to be in the loser group (i.e., bottom tercile of performance). Therefore, small hedge funds not only generate higher returns on average but also provide consistent performance over our sample period. Why does asset growth drive down performance over the life cycle of a hedge fund? Possible explanations from the literature include managers limited abilities, negative price impacts from large block trading, and the hierarchy cost discussed in Stein (2002). More recent studies provide another possible and testable explanation, namely that the standard compensation contract in the hedge fund industry is not effective at aligning managers incentives with investors interests. Lan, Wang, and Yang (2013) show that the present value of managers future management fees is much higher than the present value of future incentive fees. Yin (2016) shows empirically that the management fee comprises a larger portion of total compensation when funds grow large and thus a fund s optimal size from a compensation perspective exceeds the size that is optimal for performance. 6 These studies suggest that diseconomies of scale may reflect compensation arrangements that are weighted more toward management fees than incentive fees 6 Yin (2016) presents cross-sectional evidence of scale diseconomies but does not examine the time-series trend. 5
when funds grow sufficiently large. Consistent with this line of reasoning, we find that the contribution of future management fees to total future fees is higher for larger funds and that the contribution becomes even higher as funds grow larger over time. Thus, when funds grow over their life cycle, managers are likely to have diminishing performance incentives because most of their compensation comes from the asset-based management fee. Last, one other possible explanation for declining performance with age is that young funds are willing to take on more risk. If these risks pay off, young funds can attract capital inflows and ultimately collect more fees. However, our results do not support this explanation. Using measures such as VaR, expected shortfall, and tail risk, we find that younger funds do not have higher downside risk. As argued in Aggarwal and Jorion (2010), managers of young funds may have more innovative ideas and trading strategies relative to old funds. 7 Consequently, young funds might be able to generate good performance without taking extra risk. This study contributes to the hedge fund literature in the following key ways. First, our study is one of the very few that examines the performance life cycle of hedge funds. We find that, on average, hedge fund performance declines over its life cycle and that fund growth over time significantly drives this decline. We do not find that performance declines over the life cycle is associated with a variety of other fund- and family-level characteristics, nor do we find that is related to young funds assuming higher downside risk. Second, our study contributes to the performance persistence literature. Our results suggest that fund growth and diseconomies of scale contribute to the lack of performance persistence in the hedge fund industry. Thus, funds that maintain a small size may provide higher performance persistence. Third, we are the first study to examine how hedge fund managers incentives vary over time as a function of fund size. We show that the relative importance of the management fee to managers compensation arrangements increases with fund size over the life cycle. These findings provide further evidence that the standard compensation contract in the hedge fund industry does not align managers incentives with investors best interests. 7 We examine innovation over the performance life cycle in Section IV, B.5. 6
I. Data and Methodology A. Data We collect hedge fund data from the Lipper TASS and Hedge Fund Research (HFR) databases. Following the literature, we only consider funds that report monthly net-of-fee returns in US dollars (USD). Fund-month observations with missing information about fund returns, assets under management, or investment styles are deleted. We also exclude funds in the Fund of Funds style because they invest in other hedge funds rather than securities. To mitigate survivorship bias, we retain defunct funds in our sample. Because defunct fund data are available starting in 1994, our sample period begins in January 1994 and spans through December 2016. In addition, because Lipper TASS and HFR use different investment style classifications, we follow Agarwal, Daniel, and Naik (2009) and consolidate reported styles into the following four general styles: Directional Traders, Relative Value, Security Selection, and Multi-Process. To identify and remove duplicate funds across databases, we first identify management firms that report to both databases. We match management firms by name and by reported address. Within matched management firms, we calculate return correlations between funds in TASS and funds in HFR. For each pair of funds with correlation 0.999, we confirm the pair s duplication status based on fund name and fund returns. In addition, as pointed out in Aggarwal and Jorion (2010), management firms may report multiple share classes, including master-feeder structures, to a database. To eliminate duplicate share classes, we calculate the return correlation between each pair of funds within the same management firm. For each pair of funds with correlation 0.999, we retain the one with the longer performance record or with larger assets under management. As is well documented in the literature, hedge fund performance in commercial databases suffers from backfill bias. There are three key dates that are relevant to this bias: the inception date, the performance start date, and the add-date. The inception date is when the legal fund structure was established. The performance start date is the date of the first reported monthly return. The add-date is the date when a fund chooses to start reporting to a commercial database. Backfill happens when the performance start date precedes the add-date. Because hedge funds are more 7
likely to report performance data when performance is good, backfilled data are likely to be upward biased. To mitigate the impact of backfill bias on our analysis, we exclude backfilled funds from our main sample. Following Aggarwal and Jorion (2010), we define a fund as backfilled if the period between its inception date and its add-date exceeds 6 months. 8 Turning to the remaining non-backfilled funds, we require funds to have inception dates after 1994 and to have at least 5 years of data. 9 As discussed in the Introduction, funds that fail because of adverse industry shocks or bad luck are likely to experience poor performance just prior to liquidation. By excluding these funds, we mitigate a potential bias that could lead us to observe a negative age-performance relation even when there are no actual age effects on performance. Lastly, we require funds to start with at least $1 million in assets under management (AUM). We choose $1 million because it is the most common minimum investment requirement over our sample period, as shown in the Internet Appendix Table IA.II. Moreover, Internet Appendix Table IA.III shows that $1 million is at about the 25 th percentile of fund starting size over our sample period. Thus, the $1 million starting size requirement does not eliminate too many funds from our sample. Note that our results are not biased by these criteria, as shown in Section IV. To facilitate our analysis, we construct a reference sample using all backfilled funds and non-backfilled funds that do not survive the filters above. For the reference sample, we require funds to have at least $5 million in AUM, and we exclude observations before their add-dates. If the add-dates are not available, we remove the first 18 months of data. Finally, to mitigate the impact of outliers and reporting errors, we winsorize fund returns in both the main sample and the reference sample at the 0.5% and 99.5% levels. 8 One potential issue with this procedure is that the TASS database stopped updating add-date information around 2011. As a result, some funds without backfilled data are excluded from our main sample because their add dates are missing. To address this issue, we follow the procedure in Jorion and Schwarz (2017) to estimate add-dates for TASS funds with inception dates after 2011. Implementing this procedure adds only one additional fund to our main sample, probably because we require 5 years of data to be included in our main sample. Unsurprisingly, the results are robust to implementing this additional procedure (see Internet Appendix Section B.3). 9 We choose 5 years because the median life span of defunct funds in our sample is 58 months, which is approximately 5 years (see Internet Appendix Table IA.I). As an additional consideration, institutional investors commonly require hedge funds to have a three to five year performance record before investing in them. 8
B. Performance Measures In this study, we use two measures of fund performance: (1) net-of-fee raw returns, as reported in TASS and HFR, and (2) style-adjusted returns. Style-adjusted returns are defined as the difference between fund monthly returns and the average return of all funds in the same investment style. Thus, for fund i in month t, its style-adjusted return is defined as: SSSSSSSSSS-AAAAAA RRRRRRRRRRRR ii,tt = RRRRRRRRRRRR ii,tt SSSSSSSSSS RRRRRRRRRRRR ii,tt. (1) Raw returns are directly observable to all investors, and style-adjusted returns can be easily calculated from raw returns. Both measures are less noisy than risk-adjusted return measures estimated from factor models. 10 In addition, investors are likely to evaluate and compare hedge funds within the same style because funds in different styles may face very different markets and use significantly different investment strategies. Thus, the style-adjusted return is a reasonable measure of relative performance that provides a good complement to the raw return. C. Summary Statistics Panel A of Table I presents the summary statistics for our main sample. The means of afterfee returns and style-adjusted returns are 0.61% and 0.10% per month, respectively. The pooled mean fund age is approximately 60 months, or about 5 years. Average fund size is $170 million, while average fund family size is over $700 million with an average of a little over 4 funds per family. Following Sirri and Tufano (1998), we calculate fund flows as in Equation 2 below, and the average flow is 1.35% per month. Funds in our main sample charge a management fee between 1% and 2% and an incentive fee of 20%. Most funds have a high-water mark provision (mean is 0.92) and use leverage (mean is 0.69). In terms of share restrictions, lockup periods are not commonly used, and the mean is about 4 months. FFFFFFFF ii,tt = AAAAAA ii,tt AAAAAA ii,tt 1 (1+RRRRRRRRRRRR ii,tt ) AAAAAA ii,tt 1. (2) 10 While the literature commonly uses a rolling window approach to estimate risk-adjusted returns, such an approach is inappropriate in our setting because we examine hedge fund performance from a life cycle perspective (e.g., a rolling window approach would exclude data from early years when evaluating risk-adjusted returns in later years). 9
Panel B of Table I reports the summary statistics for funds in our reference group. The performance of the reference group is slightly lower than the performance of our main sample, with an average raw return of 0.49% and an average style-adjusted return of -0.02% per month. The means of fund size and family size are $263 million and $622 million, respectively. Hedge funds in the reference group charge an average management fee of 1.5% and an average incentive fee of about 20%. Most funds have a high-water mark provision, use leverage, and do not have lockup periods. [Insert Table I about here] II. Life Cycle of Hedge Fund Performance A. Performance of Hedge Funds: Event Time We begin our analysis by examining the performance life cycle of hedge funds. We follow Aggarwal and Jorion (2010) and group fund-month observations by event time, where event month 1 is the month of a fund s performance start date, event month 2 is the following month, etc. Next, we form equal-weighted portfolios of funds for each event month. We treat the first 12 event months as the first event year, the next 12 event months as the second event year, etc. Table II presents the average monthly performance for each event year. 11 The average raw return is decreasing with fund age, from 1.30% per month in the first year to 0.41% per month in year 5. After the fifth year, some funds become liquidated and the pattern is somewhat mixed. However, the performance in later years is never as good as it is during the first few years. Because the raw return results might be driven by funds in certain style categories, we also examine styleadjusted returns in the last two columns of Table II. Style-adjusted returns exhibit a similar pattern of declines, from 0.62% per month in year 1 to -0.01% per month in year 5. After year 5, most style-adjusted returns are negative, which suggests older funds underperform their style average. [Insert Table II about here] 11 We only present the first 10 event years in Table II because less than half of the funds in our main sample survive past their 10th year. The complete performance life cycle is reported in Internet Appendix Table IA.IV. 10
In columns 4 and 6 of Table II, we compare portfolio performance in adjacent years using a t-test. The results indicate that there are two significant performance declines over the first 5 event years using both raw returns and style-adjusted returns. The first decline occurs when going from year 1 to year 2. As argued in Aggarwal and Jorion (2010), newly established hedge funds may generate superior performance from implementing innovative strategies. However, performance does not remain at a high level for long, either because other funds learn about and adopt innovative strategies or because those strategies have limited capacities. 12 The second decline in performance occurs when going from year 4 to year 5. This result may reflect our chosen minimum fund life span of 5 years, as funds that liquidate in year 6 are likely to have poor performance in year 5. If so, the decline would support our earlier concern that fund failures impart a bias in favor of finding a negative age-performance relation. Overall, the results in Table II show that hedge fund performance declines with fund age, consistent with anecdotal evidence and prior studies such as Aggarwal and Jorion (2010). Performance appears to peak in year 1, significantly drops off in year 2, and gradually declines thereafter. The performance life cycle evidence in Table II may also help explain why prior studies have trouble documenting long-term performance persistence. Prior studies commonly find evidence of short-term performance persistence lasting up to six months, but persistence evidence becomes mixed when extending the performance horizon beyond one year. The results in Table II indicate that, on average, hedge fund performance declines over time, which suggests long-term performance persistence might be difficult to achieve. Nevertheless, it may be possible that certain types of hedge funds can provide consistent performance. In Sections II.B and II.C, we examine whether the patterns in Table II relate to a variety of fund or family characteristics, respectively. B. Fund Performance and Fund Characteristics As documented in the literature, hedge funds with different characteristics may have different performance. Thus, the return patterns in Table II could be driven by certain fund-level 12 We examine innovation over the performance life cycle in Section IV, B.5. 11
characteristics. In this section, we examine four specific fund characteristics. The first one is a lockup period, which is a period of time wherein investors cannot redeem their money. Because fund managers do not need to worry about redemptions during lockup periods, managers may use the time to invest in illiquid assets. Consequently, funds with lockup periods may be able to generate superior performance, especially during the early years of their life cycle. The second characteristic is a high-water mark provision, which requires managers to make up for any past losses before they can collect an incentive fee. Managers of funds with high-water mark provisions may deliver strong performance because they face strong incentives to keep their fund value above the high-water mark. The third characteristic is the incentive fee percentage. Although most hedge funds charge a 20% incentive fee, some funds deviate from the industry standard, and such deviations may create different incentives for performance. The last characteristic is fund leverage, which allows managers to take on additional risk to boost performance. To examine the impact of these fund characteristics, we divide our sample into subgroups and conduct the event-time analysis as in Section II.A. The results are summarized in Table III, and for simplicity, we only present results for the first 5 event years. [Insert Table III about here] In Panel A of Table III, we divide our main sample into two groups based on whether or not they have lockup periods. The results indicate that fund performance in both groups decreases with fund age. The last two columns compare the performance of the two groups using a t-test. The results show that while funds with lockup periods tend to have better performance over the first 5 years, the performance gap is generally not statistically significant. Table III, Panel B examines the impact of the high-water mark provision. Using raw returns, we find that fund performance declines over time for funds with and without high-water mark provisions and that performance differences between the groups are not statistically significant. Using style-adjusted returns, it is interesting to see that the performance of funds without highwater marks decreases to around zero after the first year. By contrast, funds with high-water marks maintain reasonable performance over their first four years. However, performance differences are not statistically significant. 12
In Panel C of Table III, we divide hedge funds into three groups based on their incentive fee percentages. Using both raw and style-adjusted returns, we find that the performance of funds with incentive fee percentages at or below 20% decreases over time. However, we do not observe a clear pattern for funds that charge an incentive fee higher than 20%. For instance, funds with incentive fees greater than 20% underperform funds with incentive fees less than 20% in the first event year, and the difference is statistically significant using raw returns. However, funds with incentive fees greater than 20% significantly outperform their lower fee peers in event years 2 through 4 using style-adjusted returns. Panel D of Table III compares hedge funds with and without leverage. In both groups, we continue to observe declining performance with age. The results of the t-tests in the last two columns indicate that funds with leverage deliver better performance than funds without leverage in most years, although superior performance is only significant in event year 2. Overall, the results in Table III indicate that hedge funds with varying fund-level characteristics generate superior performance early in their lives, only to see performance decline over time. Although we find that funds with lockup periods, high-water mark provisions, and leverage typically generate higher performance, performance differences are generally not statistically significant. C. Fund Performance and Fund Family Characteristics In addition to fund-level characteristics, fund family-level characteristics may also influence hedge fund performance. Fung et al. (2016) argue that fund families have strong incentives to generate good performance for their flagship funds (i.e., their first funds) and use the flagship s performance record to attract capital flows and launch new funds. Therefore, the outperformance of young funds may be stronger among flagship funds. Moreover, Boyson (2008a) shows that fund families that focus on their core competencies have better performance, implying that non-flagships funds that share the flagship s investment style may also outperform. Table IV reports the results of our tests of these hypotheses. In Panel A, we divide our sample into flagship funds and non-flagship funds. Following the literature, we define a fund as a 13
flagship fund if it is the first fund established by its family. The results show that funds in both groups provide superior performance in their early years. Notably, flagship funds outperform nonflagship funds in all five years. The differences are statistically significant in the first event year, and the differences in raw returns and style-adjusted returns are 0.35% and 0.28% per month, respectively. The performance differences are still economically significant over event years 3 to 5 but not statistically significant at conventional levels. In Panel B, we further divide the nonflagship funds into two groups based on whether they use the same investment strategy as their family s flagship fund or not. Again, we find that performance declines with age for both groups. Interestingly, the t-tests in column 3 suggest that the performance of non-flagship funds that use the same investment strategy as their family s flagship fund decreases at a slower pace than other non-flagship funds. In addition, they outperform other non-flagship funds in most years, although performance differences are mostly insignificant. [Insert Table IV about here] To summarize, the return patterns in Table II do not appear to be driven by funds with certain family-level characteristics. Younger hedge funds provide superior performance, and fund performance declines with fund age. Meanwhile, we also find that flagship funds outperform nonflagship funds, and that non-flagship funds employing the same investment strategy as their family s flagship fund outperform other non-flagship funds. However, the performance differences are not statistically significant in most cases. III. Fund Size and Performance Declines with Age In Section II, we find that hedge fund performance declines as funds age and funds generate superior performance at the early stage of their life cycle. Further analysis reveals that performance declines with age are not limited to funds with certain fund- and family-level characteristics. So what could be driving the decline in performance with fund age? One possible explanation is that fund growth over time erodes performance. As shown in Internet Appendix Table IA.IV, both the mean and median of fund assets increases with fund age. Prior studies, such as Teo (2009), Getmansky (2012), and Yin (2016), show in cross-sectional settings that hedge funds suffer from 14
diseconomies of scale, that is, performance decreases with fund size. In this section, we examine whether fund growth contributes to performance declines with age in the hedge fund industry. A. Fund Size and Diseconomies of Scale Prior studies commonly use panel data to examine the impact of fund size on fund performance and thus look at the size-performance relation in the cross-section. In this study, we complement the literature by examining diseconomies of scale in the time-series dimension. The time-series dimension is more appropriate for our study because we are interested in examining whether diseconomies of scale contribute to performance declines over the life cycle of a hedge fund. To this end, we employ two different approaches. For our first approach, we form size portfolios based on assets under management (AUM) at the beginning of each event year, and we divide funds into three groups using two fixed cutoff points: $10 million and $100 million. If we had instead assigned size classifications based on inception year size or cohort size (e.g., Boyson (2008b) and Aggarwal and Jorion (2010)), we might classify as small funds that grow to be quite large in their later years, while we might classify as large funds with starting sizes that are small relative to older funds in our sample. We chose $10 million and $100 million as our cutoff points based on the size distribution shown in Internet Appendix Table IA.V. For each size group, we form an equal-weighted portfolio, and we hold the portfolio over the event year. Table V reports the average performance of each portfolio over time. [Insert Table V about here] The results show that hedge funds in all three size groups generate superior performance in their early years, but that performance decreases with age. However, funds in the small size group suffer milder declines with age relative to the medium and large size groups, and the small group outperforms the large group over multiple event years. For example, we find that the yearto-year decrease in performance for small funds is mostly insignificant based on t-tests and that style-adjusted returns for small funds are positive in each event year. By contrast, performance of large funds decreases significantly after the first year, and their style-adjusted returns drop to nearly zero. Moreover, when we compare performance between the small and large groups in the 15
last two columns, we find that small funds outperform large funds in event years 2 through 5, with statistically significant outperformance in years 2 through 4. Overall, the results suggest that the deterioration of fund performance is driven primarily by large hedge funds, while small hedge funds generally maintain good performance. One drawback to the portfolio approach is that portfolio membership varies across event years as fund size changes over time. To better distinguish size and age effects, our second approach uses a modified version of the Fama-MacBeth regression. Specifically, following Coval and Shumway (2005), we first perform time-series regressions for each fund, and we then take cross-sectional averages of the fund-specific coefficients and use the averages as our estimates of the size and age effects on performance. We use this modified regression for two reasons. The first reason is that we are interested in how size and age influence performance over the life cycle of a hedge fund. Thus, conducting time-series regressions for each fund is more appropriate for answering our research question. The second reason is that we have a large number of funds that exist only for a short period of time. 13 The results of the modified Fama-MacBeth regressions are reported in Table VI. [Insert Table VI about here] In Panels A and B, the dependent variables are raw returns and the style-adjusted returns, respectively. In regression (1), we only include fund age as the independent variable. The coefficients are negative and significant in both panels (coefficients are -0.40 and -0.29, respectively, and t-statistics are -19.04 and -15.10, respectively). These results are consistent with Table II and suggest that fund performance declines with fund age. In regression (2), we only include fund size as the independent variable. The coefficients are also negative and significant in both panels (coefficients are -1.18 and -0.61, respectively, and t-statistics are -13.97 and -12.31, respectively). These results are consistent with Table V and consistent with the diseconomies of scale documented in the literature. In regression (3), we include both fund size and fund age as independent variables. In Panel A, the coefficient on size is -1.12 with a t-statistic of -14.81, while 13 See Skoulakis (2008) for more details regarding the econometrics of the modified Fama-MacBeth regression. 16
the coefficient on age is -0.07 with a t-statistic of -1.76. In other words, after controlling for fund size, the impact of fund age on performance diminishes precipitously in magnitude and becomes only marginally significant. We find similar results using style-adjusted returns in Panel B. In regression (4), we add controls for the number of other funds in the fund family, total assets of other funds in the fund family, and fund capital flows. In both panels, the coefficients on size are negative and highly significant, but the coefficients on age become insignificant. Economically, a 10% increase in fund size in regression (4) is expected to result in a decrease of 13 basis points per month (or 1.53% per year) in raw returns and a decrease of 10 basis points per month (or 1.21% per year) in style-adjusted returns, holding all other variables in the regression constant. The results in Tables V and VI suggest that hedge funds suffer from diseconomies of scale and that fund growth over time significantly contributes to performance declines as funds age. Our results also speak to the persistence of hedge fund performance. On the one hand, the literature documents that investors chase fund performance. Thus, good performance attracts capital inflows and fund growth erodes fund performance. Eventually, as predicted by Berk and Green (2004), all superior performance will be chased away, leaving no persistence in fund performance. On the other hand, our results imply that hedge funds can maintain good performance if they can restrict fund growth. Given that many hedge funds close themselves off to new investment, restricting fund growth is quite feasible in practice. B. Implications for Investors Fund Selection So far, our results suggest that the decline of hedge fund performance with age is related to fund size and that smaller hedge funds are able to generate and maintain superior performance for multiple years after their inception. Thus, a natural question is whether investors could profitably exploit these results in real time using the universe of hedge funds. To help answer this question, we pool our main sample and our reference sample and assign each fund to one of nine (3x3) portfolios based on size (small, medium, and large) and age (young, mid-age, and old) at the beginning of each calendar year. Along the age dimension, we define young funds as those that are no more than 2 years old, old funds as those that are at least 5 years old, and mid-age funds 17
as those in between. Along the size dimension, we use $10 million and $100 million as the cutoff points. Thus, portfolios are formed on independent sorts of size and age. Then we form an equalweighted portfolio for each group and hold the portfolio for one year. [Insert Table VII about here] Table VII reports the average performance of each portfolio over our sample period. Panel A shows the average raw return of each portfolio. When we compare portfolio performance along the size dimension, we find that small funds outperform large funds across all three age groups. The differences by size are statistically significant for the young and mid-age groups. By contrast, when we compare portfolio performance along the age dimension, funds in the young group only outperform the old group when fund size is small, and the young-old differences are statistically insignificant, regardless of size. Patterns are similar in Panel B, where performance is measured using style-adjusted returns. Small funds provide superior performance in all age groups, while young funds do not always outperform funds in the old group. The results in Table VII suggest that selecting smaller funds might be a good strategy for investors. Although we find that small funds, on average, outperform medium and large funds in all age groups in Table VII, these results do not indicate whether small funds provide superior performance consistently over our sample period. To address this question, we rank the performance of the nine size-age portfolios in each calendar year, labeling the top three portfolios as winners, the bottom three portfolios as losers, and the middle three portfolios as neutral. We then calculate the frequency with which each portfolio is classified as a winner, loser, and neutral portfolio over our sample period. The results are summarized in Figure 1. [Insert Figure 1 about here] Over our sample period, portfolios with small funds are more likely to be classified as winners than portfolios with medium and large funds. For instance, portfolios with small & young funds and small & mid-age funds are both winners in over 60% of the calendar years in our sample. This frequency far exceeds the frequencies of the other portfolios. Meanwhile, although we find that the large & young portfolio is classified as a winner more often than the 18
small & old portfolio, the small & old portfolio is classified as a loser less often than the large & young portfolio (20% for the small & old portfolio vs. 50% for the large & young portfolio). Taken together, the results in Table VII and Figure 1 indicate that small funds not only outperform medium and large funds (on average), they also generate superior performance consistently over our sample period. These results are quite useful for hedge fund investors. For example, even when investors (e.g., institutional investors) require a multi-year performance record to invest in a hedge fund, they can achieve stable returns if they invest in small hedge funds. C. Managers Incentives The literature provides several explanations for the negative effect of fund size on fund performance, including managers limited abilities, the price impact of large block trading, and the hierarchy cost discussed in Stein (2002). More recent studies provide another possible explanation related to how managers incentives change with fund size. Lan, Wang, and Yang (2013) show in their baseline model that 75% of the total value created by managers comes from the management fee. Lim, Sensoy, and Weisbach (2016) show that hedge fund managers indirect incentives per dollar change in fund value decrease with fund age. Thus, younger funds may have stronger incentives to improve fund performance. One way to interpret this is that additional capital flows become less important as funds grow larger. Yin (2016) shows that because of diseconomies of scale, the management fee becomes the more important part of a hedge fund managers total compensation in absolute dollar terms when funds grow large. Consequently, managers of large funds may have weaker incentives to deliver strong performance, as chasing performance may risk eroding fund size. Based on the above literature, we examine whether the impact of fund size on fund performance corresponds with changes in managers incentives over time. To be more specific, we calculate the present value of managers future fees at the end of each event quarter based on the baseline model in Goetzmann, Ingersoll, and Ross (2003; GIR hereafter). Note that unlike most empirical studies in the literature, which calculate realized compensation at the end of each quarter, we focus on managers expected compensation for the future. Because managers behavior 19
cannot change realized fees but can influence future compensation, our measure is more likely to capture managers incentives. We use the GIR model because it provides a closed-end solution to, and a lower bound on, the magnitude of managers future compensation. The calculation requires the market value of each investor s assets in the fund and their individual high-water marks. Because these values are not provided by commercial databases, we estimate them following the approach in Agarwal, Daniel, and Naik (2009). 14 We measure managers incentives as the contribution of the present value of future management fees to the present value of future total fees as follows: FFFFFFFFFF% = PPPP oooo FFFFFFFFFFFF MMMMMMMM FFFFFFFF PPPP oooo FFFFFFFFFFFF TTTTTTTTTT FFFFFFFF 100. (3) Panel A of Table VIII reports FMFEE% at the end of each event year. Following the literature, we assume that managers skills (represented by α) are either 0 or 3% per year and the withdrawal rate (represented by δ+λ) is either 5% or 10% per year. First, we find that the management fee comprises most of the managers total compensation in present value terms, as FMFEE% is higher than 50% over each of the first five event years. Second, FMFEE% increases over time. Because fund size also increases over time, our results are consistent with the literature and the intuition that the management fee becomes more important as funds grow large. Thus, fund managers may have lower incentives to improve fund performance because most of their compensation comes from the management fee, which only depends on fund assets. Moreover, as funds grow larger, incentives become even lower, thereby creating a self-reinforcing process. [Insert Table VIII about here] In Panels B through E, we divide our sample into three size groups (i.e., small, medium, and large). The results indicate that funds in the small group have lower FMFEE% in all event years, and the differences between the small and large groups are all statistically significant. Notice that, for small funds with skills (i.e., α=3%) in Panels D and E, FMFEE% is actually below 50%. In other words, managers of small funds collect more of their compensation from the incentive fee. However, as fund assets increase, we still find that FMFEE% grows over time. Also note that 14 The details of our calculation are outlined in the Appendix. 20